US Joins Musk’s xAI in Suing Colorado Over AI Anti-Discrimination Law

The U.S. Department of Justice joined Elon Musk’s artificial intelligence company xAI in April in suing Colorado to block a state law designed to prevent AI-driven discrimination.

When the federal government sides with a billionaire against a state trying to protect residents from AI bias, that is not only a Colorado story, a researcher wrote. It is everyone’s story.

The Justice Department’s lawsuit is part of a coordinated federal effort to reframe AI consumer protections as ideological overreach, said Genevieve Smith, a postdoctoral research fellow at Stanford University and founder of the Responsible AI Initiative at the UC Berkeley AI Research Lab.

In July 2025, President Donald Trump signed an executive order on “preventing woke AI,” equating bias mitigation measures to a leftist agenda that suppresses free speech and truth, Smith noted. The federal National Policy Framework launched in March included a push to pre-empt state AI laws, with Colorado’s law targeted.

The Justice Department’s intervention in Colorado marks the first time the federal government has sought to intervene in a lawsuit challenging a state AI law, Smith wrote in an opinion piece published by The Guardian.

The Colorado law, Senate Bill 205, aimed to protect people from discrimination in high-risk AI systems making consequential decisions: who gets hired, gets housing or receives healthcare, Smith wrote.

The original bill required bias audits, impact assessments and disclosure, which drew pushback from the business community. Colorado responded by revising those requirements in mid-March with a reduced transparency framework before xAI filed its lawsuit, Smith said.

A 2019 study published in Science found that a widely used healthcare algorithm deployed across U.S. hospitals assigned Black patients half the care of equally sick white patients because it used healthcare costs as a proxy for health needs, embedding existing inequities into clinical decisions, Smith wrote. Removing healthcare costs eliminated racial bias in the model and helped hospitals deliver more effective care, she said.

xAI argued the Colorado bill would force the company to promote the state’s “ideological views on various matters, racial justice in particular” in its Grok chatbot, Smith wrote. Colorado state Representative Brianna Titone, one of the bill’s lead sponsors, captured the misleading nature of xAI’s lawsuit, Smith wrote. “SB 205 is about consequential decisions … We’re not restricting speech. Our bill does not say that Grok still can’t be a dick,” Titone said.

The Justice Department called the bill “state-mandated discrimination” that “obligates AI developers and deployers to discriminate,” an extraordinary characterization of a law that requires companies to check whether their systems produce unlawfully discriminatory outcomes and fix them if so, Smith wrote.

The Justice Department’s argument rested on the false assumption that AI systems use “neutral criteria,” Smith wrote. Under this logic, altering models that use neutral criteria messes with their objectivity, constituting discrimination, she said. But as the hospital study shows, criteria that appear neutral can produce worse outcomes for some populations, Smith wrote. This trend of poor proxies harming marginalized groups has been shown in various studies for welfare allocation and college admissions, she said.

Bias also enters when models have less training data on some groups, causing worse performance for those populations as documented in facial recognition research and Smith’s own research on large language models, she wrote. When companies improved representation in their training data, performance discrepancies nearly disappeared, Smith said.

Then there is the “winning the AI race” framing reflected in the Justice Department lawsuit, Smith wrote. The race technologists talk about and that the U.S. government refers to is the pursuit of artificial general intelligence and beating China a frontier dominated by a handful of well-capitalized corporate labs, she said. Rules barring AI from discriminating in hiring and healthcare are separate from that race, and conflating the two is misleading, Smith wrote.

The only mainstream national outlet to cover this story thus far, the Wall Street Journal, focused on effects the law could have on businesses and “killing the entrepreneurial spirit,” a sentiment echoed in the Justice Department lawsuit, Smith wrote. That argument does not hold up either, she said. No companies are cited as leaving Colorado because of this regulation, and Governor Jared Polis states that “far more” firms are moving to Colorado than leaving, Smith wrote. The one concrete example offered in the Journal was Palantir citing the bill as a possible burden in a Securities and Exchange Commission filing, but the company did not say this was the reason it left Colorado for Florida, Smith said.

On the opposite end of the spectrum, companies including Microsoft have flagged bias and discrimination in AI as a material risk, Smith wrote. Small businesses have legitimate concerns, to which the state responded, she said. Still, using AI to make consequential decisions about who gets hired or receives healthcare is a choice, Smith wrote. If a business wants to develop or use AI in hiring or healthcare but does not have the bandwidth to check whether it discriminates, then they do not have the bandwidth to use it responsibly, she said. AI tools that do not work for everyone deliver worse outcomes, create liability and erode trust, Smith wrote. The healthcare algorithm that was fixed did not sacrifice efficiency; it got better at its job, she said.

But the pressure campaign worked, Smith wrote. On 14 May, Polis signed into law a new bill (SB 189) which repeals and replaces SB 205, she said. Gone are requirements for companies to proactively assess their high-risk AI systems for bias, conduct annual reviews, report discovered discrimination to the state and use reasonable care to address known and foreseeable harms, Smith wrote. What remains is a requirement for AI developers to share technical documentation (on intended uses, limitations, training data) with deployers but not with the public, she said. Consumers get a notice that AI was involved in a decision and the right to request human review, which most consumers will never know to invoke, Smith wrote. It is better than nothing but falls short of ensuring meaningful, proactive accountability for high-risk systems, she said.

If the Justice Department can join a billionaire’s lawsuit to kill a state law meant to protect consumers and support transparency in high-risk areas, what does that mean for other states trying to shield residents from AI harms in the absence of federal regulation? Smith asked. The message is clear: don’t try, she wrote. That is not pro-America, Smith said. AI is an incredibly powerful technology, and its potential is greatest when it works for all of us, she wrote. We deserve the transparency and protection to get there, Smith concluded.